Eric Schmidt argues against a ‘Manhattan Project for AGI’
In a policy paper published on Wednesday, former Google CEO Eric Schmidt, scale AI CEO Alexandr Wang, and Center for AI Safety Director Dan Hendrycks, said that the US does not have to pursue a Push-like Push-Stijl from Manhattan to develop AI systems with “superhuman, also known, also, also, also known,
The paper entitled “Superintelligence strategy“Claims that an aggressive offer of the US to control only super intelligent AI systems, could cause severe retribution from China, possibly in the form of a cyber attack that could destabilize international relations.
‘[A] Manhattan Project [for AGI] Supplies that rivals will agree to a permanent imbalance or omnicide instead of moving to prevent this, ”write the co-authors. “What starts when a urge for a super weapon and global control threatens to cause hostile countermeasures and escalating tensions, which undermines the strategy that the strategy claims to protect.”
Co-author of three very influential figures in the American AI industry, the newspaper comes only a few months after an American congress committee had proposed a “Manhattan project style” effort to finance AGI development, modeled on the atomic boom program of America in the 1940s. American Minister of Energie Chris Wright said that the ends “said”The start of a new Manhattan project“On AI while he stands for a supercomputer site next to OpenAi co-founder Greg Brockman.
The Superintelligence Strategy Paper challenges the idea, defended by various American policy and industrial leaders in recent months, that a program supported by the government that AGI strives is the best way to compete with China.
In the opinion of Schmidt, Wang and Hendrycks, the US is in something of an AGI distance mutually insured destruction. In the same way as global powers, not seek monopolies about nuclear weapons that can cause a preventive attack of an opponent to Schmidt and his co-authors that the US should be careful to racing to dominating extremely powerful AI systems.
Although comparing AI systems with nuclear weapons may sound extreme, world leaders AI already consider a top military advantage. The Pentagon says that AI helps to accelerate the army chain of the army.
Schmidt et al. Introduce a concept that they call Mutual Assured AI failure (MaIM), in which governments can proactively disable threatening AI projects instead of waiting for opponents to arm Agi.
Schmidt, Wang and Hendrycks suggest that the US is shifting its focus from “winning the race to super intelligence” to development methods that that Scare off other countries From creating super intelligent AI. The co-authors claim that the government must ‘expand’ [its] Arsenal of cyber attacks to disable imminent AI projects’ controlled by other countries and limit the access of opponents to advanced AI chips and open source models.
The co-authors identify a dichotomy that has taken place in the AI policy world. There are the ‘Doomers’, who believe that catastrophic results of AI development are a foregone conclusion and argue for countries that delay AI preliminary output. On the other hand, there are the ‘ostriches’, who believe that nations should accelerate the development of AI and in essence simply hope that it will all work.
De Paper proposes a third way: a measured approach to the development of AGI that prioritizing defensive strategies.
That strategy is especially remarkably from Schmidt, who has previously been pronounced about the need for the US to aggressively compete with China in developing advanced AI systems. Only a few months ago Schmidt released an OP-ED and said Deepseek marked a turning point in the AI race of America with China.
The Trump government seems to die to move forward in the AI development of America. However, as the co-authors notice, the decisions of America around AGI do not exist in a vacuum.
While the world looks at how America demands the limit of AI, Schmidt suggests and his co-authors may be wiser to follow a defensive approach.
